NEW YORK – When the polls close on November 2, SUNY Stony Brook professor Helmut Norpoth will be monitoring the results with the intensity of a high-stakes gambler watching cards turn over.

Mr. Norpoth recently predicted that President Bush would beat John Kerry in November with 54.7% of the popular vote.

A political scientist, a scholar of presidential elections, and researcher of public opinion, Mr. Norpoth says he’s not guessing. He is a member of a cadre of academics who have staked their professional reputation on the outcome of the 2004 race.

Mr. Norpoth and about eight others have each developed a political forecasting model that allows them to plug in a special formula of “predictor” data, such as poll numbers, and economic indicators, primary election results and get back an impossibly precise prediction.

Political forecasting using quantitative models is a relatively new field in academia, dating back about 25 years to when scholars had accumulated enough nationwide poll data from previous elections to begin to see patterns.

While most academics are satisfied with coming up with broad theories to explain the past, political forecasters are one of the few to leave themselves so vulnerable to be proven wrong.

“It is an area where you have to put up or shut up, where you are really tested,” said James Campbell, a political scientist at SUNY Buffalo who published his first forecasting model in 1990. He claims to have the ear of Bush adviser Karl Rove.

The forecasters’ reputation took a hit after the 2000 election when all of them predicted an easy Gore victory. The mistake they made, they say, is that they had predicted voters would give Al Gore more credit for the health of economy.

This time around, forecasters seem to be projecting a Bush victory. Mr. Norpoth’s model rests on the idea that primary election results matter. “Yogi Berra might have said it: the best predictor of an election is, well, an election,” he wrote in an upcoming article in the journal PS: Political Science & Politics.

Essentially, he argues that the more competitive the primary race is for the incumbent party, the worse the winning nominees will do in the general election.

His model would have predicted a loss for George H. W. Bush, who faced a tough primary challenge from Pat Buchanan in 1992. Bill Clinton did not face serious opposition in the 1996 Democratic primaries and then soundly defeated Bob Dole.

Mr. Norpoth said his model picks the winner of the popular vote in every election from 1912 to 2000 except for in 1960, when Kennedy beat Nixon by a little more than 100,000 votes.

A good political forecasting model ought to be able to offer a prediction well in advance of an election, he said. It also should rely on as few variables as possible and shouldn’t contradict the results of previous elections.

Mr. Campbell’s model uses Gallup Poll data available at Labor Day and the second-quarter growth rate in the GDP.

His model has predicted popular vote results for presidential elections since 1948 with an average error of 1.77 percentage points.

Michael Lewis-Beck, a political scientist at the University of Iowa who is regarded as the dean of political forecasters, last month announced that the 2004 election would be too close to call.

Using July Gallup Poll data, first-and-second-quarter economic growth indicators, and the number of jobs created over the course of the Bush administration, Mr. Lewis-Beck said it is “very possible” that Mr. Bush could win the popular vote but lose the electoral college vote.

Harvard economist Ray Fair’s model plugs in economic growth and inflation numbers and a “good news” variable that takes into account the number of quarters since Mr. Bush took office when the growth rate of real per capita GDP was greater than 3.2%.

The result? A Bush landslide with 57.5% of the vote.

“I can’t cheat. It’s out there,” Mr. Fair said.

“There is nothing I’m going to do between now and November.”